Rate how accurately the tools' own reports of success or failure matched the actual outcomes of their work

The mean rating was 4.6 and the median 5.0 (n = 786 respondents). Ratings of 3 or below, the distressed range, made up 31.93% of the answers. The rating was associated with all four study measures. Higher ratings came with less cancellation (r = −0.66, 95% CI −0.74 to −0.56, p < 0.001, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.66, 95% CI 0.53 to 0.76, p < 0.001, n = 94 engagements), a higher engagement indicator mean computed without the indicator (r = 0.87, 95% CI 0.83 to 0.91, p < 0.001, n = 150 engagements) and lower intent to leave (r = −0.63, cluster bootstrap 95% CI −0.68 to −0.57, p < 0.001 with standard errors clustered by engagement, n = 739 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 47.72, Holm p < 0.001, omega squared 0.427), respondent role (Welch F 6.24, Holm p = 0.015, omega squared 0.012) and vendor type (Welch F 4.92, Holm p = 0.002, omega squared 0.026).

Of the 73 written responses, 63 were negative and 10 positive, and 15 carried the strongest negative score of −10. Respondents described tools that reported runs, cases and checks as complete when they were not, so that teams counted, sampled and re-opened the work by hand. Across engagements, a more negative engagement mean written score came with cancellation, and respondents whose written responses were more negative reported higher intent to leave.

Ratings of 3 or below, the distressed range, accounted for 251 of the 786 answers (31.93%). The engagement mean fell in the distressed range in 35 of 150 engagements.

The primary responses numbered 62, all written in answer to C177, “Add any comments on the agentic AI tools that the ratings in this section do not capture.” The remaining 11 responses, 15.07% of the total, were secondary and came from 6 other questions. That secondary share ranked 39th from the highest among the 71 Part 2 indicators with 30 or more written responses, against a median of 18.92%. Respondents who wrote both a primary and a secondary response numbered 0. Negative responses made up 85.48% of primary responses (53 of 62) and 90.91% of secondary responses (10 of 11).

Fifty-five responses described the tool reporting a run, case, check or task as complete or successful when it was not, or stopping without an error. The examples ranged across identity, reporting, data, consent, document retrieval and reconciliation work: a recertification run reported complete on four hundred accounts when about thirty entitlements had not been revoked, a batch of disclosures marked complete with three composites unbuilt, a data load reported complete with half the rows missing, consent records reported updated that never went through, an agent reporting success with no documents pulled for two days after a portal added a login step, and a forty-thousand-dollar break matched to the wrong account and marked closed over a weekend.

Two episodes reached senior audiences. A batch of eighty investigations went to the financial institution's financial crime committee with the tool's run summary showing all eighty complete, when six had stopped part way. The respondent said explaining to the head of compliance why a machine's account of its own work is not evidence set the programme back a month. In another engagement a forecast pack, taken into a review with the finance director on the strength of a run report that showed every step complete, cited a source that did not contain the figure, and the pack was withdrawn in the room.

Four responses described dashboards, trackers or reports that used the agent's own confidence or completion figures as the measure of quality: a compliance dashboard that counted calls compliant on the agent's confidence score, a diligence tracker whose quality metric was the agent's own extraction confidence, and a data room shown as ninety percent reviewed on the agent's self-reported figure. Eleven responses described manual counts, samples or checks kept because the status could not be trusted, from counting rows each morning to tying cash before sign-off, and four said the team stopped trusting what the tool produced.

The ten positive responses said the tool reported its failures honestly, several by contrast with an older tool or an earlier year, and one said a resolved case could now be taken as resolved and checked by sample.

Of the 73 responses, 63 (86.30%, 95% CI 77.78% to 93.51%) were negative and 15 (20.55%) carried the strongest negative score of −10. Positive responses numbered 10 (13.70%), and no response was neutral. The mean written score was −4.7 (95% CI −5.6 to −3.6, n = 73 responses), and the median was −5.0.

Of the 73 responses, 71 came from respondents who also rated the indicator. Among them, the written score rose with the rating (Spearman rho 0.51, 95% CI 0.29 to 0.68, p < 0.001, n = 71 responses). The written responses corroborated the rating. Of the 61 negative responses with a rating, 40 (65.57%) came from respondents whose rating was in the distressed range.

Respondents who wrote about the indicator and gave a rating numbered 71, with a mean rating of 3.7. The 715 respondents who rated the indicator without writing about it had a mean of 4.7. The difference, −1.0 (95% CI −1.5 to −0.4), excluded zero. Ratings in the distressed range made up 57.75% of the ratings from respondents who wrote (41 of 71) and 29.37% of the ratings from respondents who did not (210 of 715).

The respondents who wrote came disproportionately from the distressed end of the scale. The written evidence therefore gives more weight to the respondents who rated the indicator worst.

Every figure in the segment tables is at respondent level. Segments with fewer than 30 respondents carry no interval. Segments with fewer than 10 report counts only.

Segment families recorded at engagement level were tested on engagement means. Respondent role and vendor type were tested on respondents. Where an expected count fell below 5, the test of the distressed share used Fisher's exact test for a two-by-two table, marked exact, or a Monte Carlo test, marked simulated. After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 47.72, Holm p < 0.001, omega squared 0.427). Active engagements in Build-Out rated highest (mean 5.5, n = 164 respondents) and canceled engagements lowest (mean 3.3, n = 296 respondents). Ratings also differed by respondent role (Welch F 6.24, Holm p = 0.015, omega squared 0.012). Portfolio directors rated highest (mean 5.4, n = 31 respondents) and individual contributors lowest (mean 4.5, n = 603 respondents). Ratings also differed by vendor type (Welch F 4.92, Holm p = 0.002, omega squared 0.026). Consulting firms rated highest (mean 5.3, n = 106 respondents) and BPO specialists lowest (mean 4.2, n = 183 respondents). Outsourcing model, institution type, process family and class of work showed no difference.

Written sentiment differed by engagement status and lifecycle stage across the 58 engagements with a written response (Kruskal-Wallis H 15.49, df 3, p = 0.001). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean −7.0, 27 responses) and responses from active engagements in Transition/Cut-Over the least negative (mean −2.4, 19 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 52, engagement managers wrote 16 and portfolio directors wrote 5 responses.

Cancellation rate is the share of engagements canceled by the study date, measured at engagement level (150 engagements, 56 canceled). Canceled engagements were rated after cancellation. The association is therefore a descriptive contrast and not a forecast of cancellation.

Across the study set, higher ratings came with less cancellation (r = −0.66, 95% CI −0.74 to −0.56, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in fifteen of the 20 segment levels tested.

Customer satisfaction is the percentage recorded against each engagement. The primary estimate uses the 94 active engagements. A descriptive contrast uses all 150 engagements.

Across the active engagements, higher ratings came with higher customer satisfaction (r = 0.66, 95% CI 0.53 to 0.76, Holm p < 0.001, n = 94 engagements). The association held after Holm correction in ten of the 16 segment levels tested.

The engagement indicator mean is the mean of an engagement's ratings across the rated indicators, measured at engagement level (150 engagements). The indicator is one of the rated indicators. Each comparison therefore uses the mean recomputed without it.

Higher ratings came with a higher engagement indicator mean (r = 0.87, 95% CI 0.83 to 0.91, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in twenty-four of the 24 segment levels tested.

Intent to leave is the answer to C019, “How likely are you to leave your current job in the next 12 months?”, rated 1 to 10. It is measured at respondent level on up to 798 respondents, with the engagement-level form reported alongside.

Respondents who gave higher ratings reported lower intent to leave (r = −0.63, 95% CI −0.67 to −0.58, Holm p < 0.001, n = 739 respondents). The association held after Holm correction in thirty-six of the 37 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 99 other engagements, 92 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 51 engagements with a negative response, 23 (45.10%) were canceled, against 33 of the 99 other engagements (33.33%). The difference did not reach significance (Fisher exact p = 0.212). Customer satisfaction among active engagements did not differ between the two groups (mean 62.00% on 28 engagements against 64.98% on 66, Welch t −1.11, p = 0.272).

Two correlations test written sentiment directly. Across the 58 engagements with a written response, a more negative engagement mean written score came with cancellation (point-biserial r −0.46, p < 0.001). Across 70 respondents, a more negative respondent mean written score came with higher intent to leave (Spearman rho −0.42, p < 0.001). Of the four comparisons, two reached significance at 0.05.

The four associations were compared in pairs on their engagement-level forms, each oriented so that a higher value is the better outcome, across the 150 engagements with all four measures. After Holm correction, five of the six pairs differed. The association with the engagement indicator mean was the strongest and the association with intent to leave the weakest. The associations with customer satisfaction and intent to leave did not differ (Holm p = 0.343).

Answers clustered within engagements (ICC 0.55). Clustering reduced the 786 respondent answers to the equivalent of about 235 independent answers. Respondent-level intervals therefore used the cluster bootstrap. Respondent-level tests of intent to leave used standard errors that allow for clustering by engagement.

No association was equivalent to zero. Each of the four differed from zero. The detection floor shown is the study floor at 150 engagements. Intent to leave was tested on its effective sample size.

Each change was computed before rounding and may differ by 0.01 from the difference between the two printed values.

All 22 checks held. A check holds when its test falls on the same side of 0.05 as the primary test. Where the primary test falls below 0.05, the check must also keep the sign of the primary correlation. The thin-base check removes the engagements in any level of outsourcing model or process family with fewer than 30 engagements. Institution type is not used for that check, because every one of its levels has fewer than 30 engagements.

Each of the four measures was computed from data the indicator does not contain. Each result is therefore a finding and not an arithmetic consequence. The indicator is one of the rated indicators, and the engagement indicator mean was recomputed without it. The correction changed r by 0.00, from 0.88 to 0.87.

Each deviation was computed before rounding and may differ by 0.01 from the difference between the two printed values.

The comparison covered 150 engagements. The indicator followed the study's typical pattern on all four measures, and no deviation held after Holm correction.

Among the 71 Part 2 indicators with 30 or more written responses, the indicator's share of negative responses, 86.30%, ranked 22nd from the highest, against a median of 80.82%. Its mean written score, −4.7, ranked 38th from the most negative, against a median of −4.7.

The indicator most often present in the same responses was C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.”, in 21 of the 73 responses (28.77%). C160, “Rate the end-to-end reliability of the tools on multi-step tasks, from first step to finished output.” (16 responses), C176, “Rate the quality of the tools' logs, traces, and usage telemetry for detecting and diagnosing failures.” (12 responses), C168, “Rate how consistently the agents took only intended and authorized actions when operating connected tools and systems.” (9 responses) and C158, “Rate the accuracy of citations, references, and quotations the tools produced against the real sources.” (7 responses) followed.

The most frequent theme was “Occurs every cycle or most days”, in 10 responses (13.70%). The second was “Comparison to a prior engagement”, in 7 responses.

About a third of all ratings of how accurately the tools reported their own success or failure fell in the distressed range (251 of 786 answers, 31.93%). The rating was associated with all four study measures, and every association held under each of the 22 robustness checks. Canceled engagements rated far below the three active stages (mean 3.3, n = 296 respondents).

The written evidence identifies false completion as a failure that defeats the controls built on top of it. When a run reports success regardless of outcome, every downstream status, dashboard and committee paper inherits the error, and the only remaining control is a person re-doing the check. Where the vendor's own reporting used the agent's self-assessment as its quality metric, that error reached the financial institution as assurance. Written sentiment was associated with cancellation and with intent to leave.

The evidence is the reported experience of the vendor's delivery teams, collected in a single cross-section. The associations reported here are not causal.

Tests by respondent role and by vendor type treated respondents as independent. The section on clustering reports how far clustering within engagements reduced the information in respondent-level figures.

Customer management executives gave 2 answers, a base of anecdote strength.

Shares of written sentiment are shares of responses and not of the study set of respondents. The written score takes five fixed values, so its mean and median are coarse. Responses clustered within engagements (73 responses on 58 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.